nablatensor

Docs / nablatensor-tensor / com.nablatensor.tensor

final class

BackendRegistry

Discovers ComputeBackend implementations and resolves them by selector/device.

View source ↗Full module reference on javadoc.io ↗

Methods

static List<ComputeBackend> all()
static List<ComputeBackend> available()

All backends that report themselves available on this machine.

static ComputeBackend defaultBackend()

The default backend: the one named by -Dnablatensor.backend=<name> / NABLATENSOR_BACKEND if that backend is present and available, otherwise the highest-priority available backend (CUDA > Vulkan > ROCm > SIMD > CPU).

static ComputeBackend forSelector(Backend selector)
static ComputeBackend forDevice(Device device)